Documentation

The metrics withMonitoring always registers.

Registered only when the RTS collects statistics, see registerRtsMetrics.

The metric names a scrape actually exposes: the identifier starting each sample line, with the suffixes Prometheus appends to a histogram removed.

Matching whole names rather than substrings is the point: renaming a counter to, say, hydra_head_inputs_total has to be caught, and the old name is a prefix of the new one.

withMonitoring forks its server and returns immediately, so a scrape can arrive before the socket is listening.